JP6497367B2 - PLANT CONTROL DEVICE, PLANT CONTROL METHOD, PLANT CONTROL PROGRAM, AND RECORDING MEDIUM - Google Patents

Description

  The present invention relates to a plant control device, a plant control method, a plant control program, and a recording medium.

  Conventionally, chemical and industrial plants, plants that manage and control wells such as gas fields and oil fields, and their surroundings, plants that manage and control power generation such as hydropower, thermal power, and nuclear power, and environmental power generation such as solar and wind power In a plant to be controlled, a plant or factory such as a plant to manage and control water and sewage, a dam, etc. (hereinafter referred to as “plant” when collectively referred to as these), a field device called a measuring instrument or an operating device A distributed control system (DCS: Distributed Control System) in which devices and control devices that control these devices are connected via communication means is constructed, and high-level automatic operation is realized.

  In a plant system or the like constructed to realize the above-described advanced automatic operation, a supervisor called a board operator monitors the operation state of the plant. Operational status such as plant yield, operating status, and alarm occurrence status is measured by a measuring instrument such as a sensor, collected via DCS, and displayed on a monitoring device such as an operation panel or a monitor of a monitoring terminal. When the board operator recognizes an abnormality in the plant through the monitoring device or increases or decreases the yield of the plant, the board operator instructs the operator called a field operator to investigate, inspect or replace the equipment. Or adjustment instructions for operating devices such as valves.

  Further, in plant control, there is a system that provides information to an operator by using artificial intelligence using a judgment model using input information from a sensor or the like as sample data (see, for example, cited references 1 and 2).

JP 2014-174993 A Special table 2015-530652 gazette

  However, in the plant operation by the board operator and the field operator, costs such as labor costs are necessary, and human error may occur.

  In addition, in a system using artificial intelligence, there is a possibility that an operating condition that does not ensure the safety of the plant is instructed to the operating device or the like. Also, in order to ensure the safety of the plant, do not input the operation instructions output by the system directly to the plant equipment, notify the worker as advice, and confirm the content of the advice from the worker. There was a case of actual operation.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a plant control device, a plant control method, a plant control program, and a recording medium that are highly safe and can reduce costs.

  (1) In order to solve the above-described problem, the plant control apparatus of the present invention includes a simulated dangerous state registration unit that registers a simulated dangerous state that simulates an operational state in which the plant is dangerous, and the operational state of the plant. An operation state acquisition unit that acquires the operation state, a learning unit that learns the acquired operation state and the registered simulated dangerous state, and generates an operation model of the plant, and the acquired operation state is generated. A determination unit that determines an operation parameter of the plant based on the operation model; and an operation instruction unit that instructs the operation of the plant based on the determined operation parameter.

  (2) Moreover, in the plant control apparatus of the present invention, the learning unit learns the acquired operating state and the registered simulated dangerous state as a loss function indicating incompatibility of the operating state, and the operation The operation model is generated by associating the state with the loss function, and the determination unit calculates the loss function of the acquired operation state, and reduces the loss function based on the generated operation model. The operation parameter to be determined is determined.

  (3) Moreover, in the plant control apparatus of the present invention, the learning unit generates the operation model so that the loss function increases in the learned simulated dangerous state.

  (4) Moreover, the plant control apparatus of this invention WHEREIN: The said operation instruction | indication part instruct | indicates the driving | operation of the said plant with respect to the apparatus which adjusts the said operation state.

  (5) Moreover, the plant control apparatus of this invention WHEREIN: The said operation instruction | indication part instruct | indicates the driving | operation of the said plant with respect to the operator who operates the apparatus which adjusts the said plant.

  (6) The plant control device of the present invention further includes a target yield acquisition unit that acquires a target yield of the plant, and the determination unit performs the operation when the acquired target yield is changed. Based on the model, the operating parameters are determined so as to avoid the simulated dangerous state.

  (7) Moreover, in the plant control apparatus of the present invention, the target yield acquisition unit further acquires a set period for the target yield, and the determination unit performs the operation based on the acquired target yield in the set period. Determine the parameters.

  (8) In order to solve the above-described problems, the plant control apparatus of the present invention is an operation control unit that controls plant equipment, a manufacturing execution unit that manages the operating state of the plant control unit, and the plant becomes dangerous. A simulated dangerous state registration unit for registering a simulated dangerous state representing the operational state; an operational state acquisition unit for acquiring the operational state of the plant; and the simulated dangerous state registered with the acquired operational state Learning unit for generating an operation model of the plant, a determination unit for determining an operation parameter of the plant based on the acquired operation state and the generated operation model, and the determined An operation instruction unit that instructs the operation control unit to operate the plant based on an operation parameter.

  (9) In order to solve the above-described problem, the plant control method of the present invention includes a simulated dangerous state registration step for registering a simulated dangerous state that simulates an operational state in which the plant is dangerous, and the operational state of the plant. The operation state acquisition step of acquiring the operation state, the acquired operation state and the registered simulated dangerous state are learned, a learning step of generating an operation model of the plant, and the acquired operation state are generated. A determination step of determining an operation parameter of the plant based on the operation model, and an operation instruction step of instructing an operation of the plant based on the determined operation parameter.

  (10) In order to solve the above-described problem, the plant control program of the present invention includes a simulated dangerous state registration process for registering a simulated dangerous state that simulates an operational state in which the plant is dangerous, and an operational state of the plant. Learning processing for acquiring the operating state of the plant by learning the acquired operating state and the registered simulated dangerous state, and the acquired operating state is generated. Based on the operation model, a determination process for determining an operation parameter of the plant and an operation instruction process for instructing an operation of the plant based on the determined operation parameter are executed by a computer.

  (11) In order to solve the above problems, the recording medium of the present invention includes a simulated dangerous state registration process for registering a simulated dangerous state that simulates an operational state in which the plant is dangerous, and an operational state of the plant. The operation state acquisition process to be acquired, the acquired operation state and the registered simulated dangerous state are learned to generate an operation model of the plant, and the acquired operation state is generated. A plant control program is recorded that causes a computer to execute a determination process for determining an operation parameter of the plant based on the operation model and an operation instruction process for instructing an operation of the plant based on the determined operation parameter. .

  According to the present invention, it is possible to provide a plant control device, a plant control method, a plant control program, and a recording medium that are highly safe and capable of reducing costs.

It is a figure which shows the structural example of the plant using the plant control apparatus in embodiment. It is a block diagram which shows an example of the hardware constitutions of the plant control apparatus in embodiment. It is a block diagram which shows an example of the software structure of the plant control apparatus in embodiment. It is a figure which shows an example of the simulation dangerous state learned with the plant control apparatus in embodiment. It is a figure which shows an example of the history data learned with the plant control apparatus in embodiment. It is a figure which shows an example of the space containing the simulation dangerous state learned with the plant control apparatus in embodiment. It is a flowchart which shows an example of operation | movement of the plant control apparatus in embodiment. It is a figure which shows an example of the target yield setting screen of the plant control apparatus in embodiment.

  Hereinafter, a plant control device, a plant control method, a plant control program, and a recording medium according to an embodiment of the present invention will be described in detail with reference to the drawings.

  First, the outline | summary of the plant using a plant control apparatus is demonstrated using FIG. Drawing 1 is a figure showing the example of composition of the plant using the plant control device in an embodiment. In FIG. 1, a plant 100 includes a plant control device 1, a core business system 2, a manufacturing execution system 3, an operation control device 4, an operation panel 5, a maintenance device 6, a field operator terminal 7, and a plant device P0.

  The plant equipment P0 includes a reactor P1, a tank P2, sensors S1 to S6, and valves V1 to V3. The plant equipment P0 generates a predetermined product (product). The reactor P1 is a device that generates a product by, for example, chemically reacting charged materials, and the tank P2 stores the product generated in the reactor P1. The sensors S <b> 1 to S <b> 6 are input devices that input signals of physical quantities (pressure, temperature, etc.) such as a differential pressure gauge, a thermometer, and a flow meter to the operation control device 4. The sensor S6 is a measuring device that measures the yield of the product generated by the plant equipment P0. In the present embodiment, the plant 100 is controlled so that the yield of the product measured by the sensor S6 becomes the target yield. The valves V <b> 1 to V <b> 3 are output devices that output a valve opening instruction from the operation control device 4, and vary the flow rate or pressure of the material or product depending on the valve opening. In the plant equipment P0, the sensors S1 to S6 and the valves V1 to V3 are hereinafter referred to as “field equipment”. The field device is a control target of the plant control apparatus 1, and by controlling the field device, the yield of the product by the plant device P0 can be controlled. In the following description, the plant equipment P0 refers to one or a plurality of equipment included in the plant equipment P0. Moreover, the plant equipment P0 illustrated in FIG. 1 illustrates the configuration of the plant, and the input device and the output device are not limited to the above configuration. For example, the input device may include switches and the like. Further, the output device may include an actuator such as a pump and a device such as a heater. Moreover, the line which connects each apparatus shown in FIG. 1 has shown the wired or wireless communication line. Wired communication or wireless communication may be performed via a communication device and a network (not shown).

  The core business system 2 is, for example, an ERP (Enterprise Resource Planning) system for the process manufacturing industry for managing management resources such as accounting processing, production management, and sales management. The core business system 2 may use information on the operation state of the plant as management information for management resources. Further, the core business system 2 may include a maintenance management system for managing business information for plant maintenance and repair. The core business system 2 is a general-purpose computer such as a server device or a desktop PC, for example.

  The manufacturing execution system 3 is, for example, a MES (Manufacturing Execution System) located between the core business system 2 and the operation control device 4. The operation state of the plant equipment P0 acquired by the operation control device 4 and the work of the operator Monitor or manage the situation. The manufacturing execution system 3 communicates with the plant control device 1 and outputs information such as a target yield acquired from the core business system 2 to the plant control device 1, for example. Moreover, the information of the driving | operation instruction | indication for operating the plant apparatus P0 acquired from the plant control apparatus 1 is acquired. The manufacturing execution system 3 is a general-purpose computer such as a server device or a desktop PC, for example.

  The operation control device 4 controls the operation of the field device by acquiring the measured value in the sensor from the input devices such as the sensors S1 to S6 and outputting an instruction for operating the output devices such as the valves V1 to V3. . In the present embodiment, the measured values acquired from the sensors S1 to S6 and the output values output to the valves V1 to V3 and the like are referred to as numerical values indicating the operation state of the plant (hereinafter referred to as “numerical indexes”). The operation control device 4 outputs a numerical index to the plant control device 1. The operation control device 4 is, for example, a device such as an FA (Factory Automation) computer or a PLC (Programmable Logic Controller).

  The operation panel 5 is a device for a field operator of the plant to monitor the operation state of the field device and operate the field device. The operation panel 5 includes display devices such as lamps and displays, or operation devices such as push button switches and a keyboard. In the present embodiment, the field operator who has received an operation instruction for instructing the operation of the plant output from the plant control apparatus 1 described later operates the field device of the plant device P0 using the operation device of the operation panel 5. And

  The maintenance device 6 is a device for the field operator to maintain the field device. Field device maintenance includes, for example, processing for reading and confirming device information set in a field device, processing for setting new device information (parameter) for a field device, and device information set for a field device. For example, a process of adjusting or changing, a process of setting device information in a field device, and executing a predetermined operation. The maintenance device 6 has a communication function for performing communication with the field device using, for example, wired communication or wireless communication. The maintenance device 6 maintains field devices using a communication function. Information transmitted and received with the field device in the maintenance performed by the maintenance device 6 using the communication function is referred to as “maintenance information”. The maintenance information may include text information, image information, audio information, and the like recorded in the maintenance device 6 by the field operator in addition to the information read from the field device as described above. The maintenance device 6 transmits maintenance information to the plant control device 1. The maintenance device 6 is a notebook or tablet computer, a PDA (Personal Digital Assistant), a smartphone, or the like.

  The field operator terminal 7 is a terminal device possessed by the field operator. The field operator terminal 7 acquires an operation instruction that instructs operation of the plant that is output from the plant control device 1. The field operator terminal 7 obtains an operation instruction from the plant control device 1 through communication means such as e-mail, chat, and voice call, and notifies the field operator. The field operator terminal 7 is a notebook or tablet computer, PDA, smartphone or the like.

  The plant control device 1 communicates with the manufacturing execution system 3, the operation control device 4, the maintenance device 6 and the field operator terminal 7. The plant control device 1 acquires the state of the field device from the operation control device 4 and the maintenance device 6. In addition, the plant control device 1 acquires a simulated dangerous state from the manufacturing execution system 3 that simulates an operation state in which the plant is dangerous. The plant control device 1 learns the operation state of the plant based on the acquired information and generates an operation model of the plant. The plant control device 1 determines the incompatibility of the current plant operation state based on the generated operation model, and determines the operation parameter to be set in the field device for optimizing the operation state. The plant control device 1 outputs an operation instruction to the operation control device 4 or the field operator terminal 7 based on the determined operation parameter. Since the plant control device 1 can output an operation instruction by substituting or assisting the board operator, human error of the board operator can be reduced and the safety of the plant can be improved. Further, since the load on the board operator can be eliminated or reduced by substituting or assisting the board operator, the running cost of plant operation is reduced by reducing the labor cost of the board operator.

  Next, the hardware configuration of the plant control apparatus 1 will be described with reference to FIG. FIG. 2 is a block diagram illustrating an example of a hardware configuration of the plant control device 1 in the embodiment.

  In FIG. 2, the plant control device 1 includes a CPU (Central Processing Unit) 11, a RAM (Random Access Memory) 12, a ROM (Read Only Memory) 13, a HDD (Hard Disk Drive) 14, a display device 15, an input device 16, It has a communication I / F (Interface) 17 and a bus 19 for connecting them.

  The plant control device 1 is, for example, a server device, a general-purpose computer such as a desktop PC, an FA computer, a device such as a PLC, a notebook or tablet computer, a PDA, or a smartphone. The plant control apparatus 1 substitutes or assists a board operator, and may be installed in the vicinity of a monitoring terminal monitored by a board operator (not shown), for example.

  The CPU 11 controls the plant control device 1 by executing a program stored in the RAM 12, the ROM 13 or the HDD 14. CPU11 runs the plant control program for implement | achieving operation | movement of the plant control apparatus 1 mentioned later in FIG. The plant control program is acquired from, for example, a recording medium that records the plant control program or a server that provides the plant control program via a network, is installed in the HDD 14, and is stored in the RAM 12 so as to be readable from the CPU 11.

  The display device 15 is, for example, a liquid crystal display having a display function. The display device 15 may be realized by various forms such as a head-mounted display, a glasses-type display, and a wristwatch-type display. The input device 16 is, for example, a keyboard or a mouse having an input function. The input device 16 may be a microphone for inputting audio information, a camera for inputting image information, or the like. The display device 15 and the input device 16 may be realized by a device having a display function and an input function, such as a touch panel.

  The communication I / F 17 controls communication with other devices such as the manufacturing execution system 3, the operation control device 4, the maintenance device 6, and the field operator terminal 7 via wired communication or wireless communication. The communication I / F 17 performs communication control such as data transmission / reception, voice call or mail transmission / reception with other connected devices. The communication I / F 17 is, for example, ISA100, which is a wireless communication standard of ISA (International Society of Automation: International Measurement and Control Society), HART (Highway Addressable Remote Transducer) (registered trademark), BRAIN (registered trademark), FOND ROD Communication control corresponding to a communication standard for industrial instruments such as the above may be performed. The communication I / F 17 may perform communication control corresponding to general-purpose communication standards such as wireless LAN communication, wired LAN communication, infrared communication, and short-range wireless communication.

  Next, the software configuration of the plant control apparatus 1 will be described with reference to FIG. FIG. 3 is a block diagram illustrating an example of a software configuration of the plant control apparatus 1 in the embodiment.

  In FIG. 3, the plant control device 1 has functions of a simulated dangerous state registration unit 101, an operation state acquisition unit 102, a learning unit 103, a determination unit 104, an operation instruction unit 105, and a target yield acquisition unit 106. Each function of the plant control device 1 is a functional module realized by a plant control program that controls the plant control device 1. The plant control program may be provided from a server that provides the program, or may be provided from a recording medium.

  The simulated dangerous state registration unit 101 acquires a simulated dangerous state that simulates the operation state in which the plant is dangerous, and registers it in a storage device such as the HDD 14 in FIG. The operation state in which the plant becomes dangerous refers to an operation state in which the failure or accident of the plant device P0 in FIG. 1 is highly likely to occur, and measurement acquired from one or more field devices such as the sensors S1 to S6. A state represented by a value. In normal plant operation, the plant is controlled so that it does not enter a dangerous operating state. Therefore, even if the plant operating state is learned based on information acquired from field devices, the plant becomes dangerous. It is difficult to learn the operating state, and such an operating state cannot be obtained in an actual plant. In the present embodiment, a simulated dangerous state in which the operational state in which the plant is dangerous is represented in a simulated manner by an index such as a measured value of a field device is prepared in advance and used for an operation model generation to be described later.

  The simulated dangerous state can be represented as a dangerous area by a numerical range of measurement values of one or a plurality of field devices. For example, when it is possible to determine whether or not the plant is dangerous based on the measurement value of one sensor, the simulated dangerous state can be represented as a dangerous area in the numerical range of the measurement value of one sensor. Further, when it is possible to determine whether or not the plant is dangerous based on the measurement values of a plurality of sensors, the simulated dangerous state can be expressed as a risk region of a combination of numerical ranges of the measurement values of the plurality of sensors. For example, in the case of a combination of measurement ranges of n sensors, the dangerous area can be expressed as an n-dimensional space. The dangerous area will be described with reference to FIG.

  In addition, the simulated dangerous state can be registered as a qualitative state that cannot be expressed by the measured value of the field device. For example, in the field device maintenance work performed by the field operator, when any abnormality is found by hearing, smell or touch, image information obtained by photographing the device, voice information recording the abnormal sound, text information input by the field operator, etc. The maintenance record can be recorded in the maintenance device 6, and the maintenance record can be put into a simulated dangerous state. The simulated dangerous state registration unit 101 may acquire and register a simulated dangerous state from the recorded maintenance information.

  The simulated dangerous state registered by the simulated dangerous state registration unit 101 is recorded in the HDD 14 of FIG. The simulated dangerous state registration unit 101 may change or delete the registered simulated dangerous state.

  The operation state acquisition unit 102 acquires the operation state of the plant equipment P0. In the present embodiment, the operating state of the plant equipment P0 is represented by the measured values of the sensors S1 to S6. For example, the operating state acquisition unit 102 requests the sensors S1 to S6 to acquire measurement values and acquires the measurement values. The driving state acquisition unit 102 sequentially records the acquired driving state.

  The learning unit 103 learns the operation state of the plant device P0 acquired by the operation state acquisition unit 102 and the simulated dangerous state registered by the simulated dangerous state registration unit 101, and generates an operation model of the plant device P0. . The learning unit 103 performs machine learning using the acquired driving state as input information. In the present embodiment, a support vector machine (SVM) technique is used as a machine learning technique. The learning unit 103 selects a measurement value to be classified as a class from the measurement values of the sensors S1 to S6. The measurement value is selected in advance. When the class to be classified is n class (multi-class), the learning unit 103 uses a multi-class SVM (MMSVM) technique in which “nC2” SVMs are prepared for the n class. By using the MMSVM method, it becomes easy to generate a model of an operation state for the measurement values acquired from a plurality of sensors in the plant.

  The learning unit 103 represents a loss function indicating the incompatibility of the operation state of the plant device P0 between the operation state of the plant device P0 acquired by the operation state acquisition unit 102 and the simulated dangerous state registered by the simulated dangerous state registration unit 101. And an operation model is generated by associating the operation state of the plant equipment P0 with the loss function. The loss function can be calculated based on the yield of the product of the plant equipment P0, for example. The instantaneous yield of the product can be increased by increasing the load such as the amount of raw material input to the plant equipment P0. However, when the load on the plant equipment P0 is increased, the operating rate may be reduced due to a failure of the plant equipment P0, and the long-term yield may be reduced. The learning unit 103 can generate an operation model by calculating as a loss function with a decrease in yield in a predetermined period as a negative bias.

  The learning unit 103 generates a driving model so that the loss function increases in the simulated dangerous state registered in the simulated dangerous state registration unit 101. For example, the learning unit 103 generates the operation model so that the negative bias of the loss function in the simulated dangerous state is maximized, and thereby the operation pattern learned so that the operation state of the plant equipment P0 does not become the simulated dangerous state. Can be generated.

  The machine learning method used in the learning unit 103 is not limited to the SVM, and for example, a neural network method such as Deep Learning may be used. The learning unit 103 generates a driving model based on the driving state acquired and recorded in the past by the driving state acquisition unit 102. The learning unit 103 may update the generated driving model when a new driving state is acquired.

  The determination unit 104 determines an operation parameter of the plant based on the operation state of the plant device P0 acquired by the operation state acquisition unit 102 and the operation model generated by the learning unit 103. Based on the target yield of the plant device P0 acquired by the target yield acquisition unit 106 and the target yield setting period, the determination unit 104 calculates a loss function for the operating state of the plant device P0 acquired by the operating state acquisition unit 102. calculate. Based on the operation model generated by the learning unit 103, the determination unit 104 determines the operation parameter of the plant device P0 so as to decrease the calculated loss function (so that the yield of the product in the set period increases). The operation parameter of the plant equipment P0 is a parameter for instructing the opening degree of the valves V1 to V3, for example.

  When the target yield is changed, the determination unit 104 recalculates the loss function based on the changed target yield, and changes the operation parameter set in the field device. Here, the determination unit 104 avoids the simulated dangerous state registered in the simulated dangerous state registration unit 101 when changing the currently set operational parameter to a new operational parameter corresponding to the changed target yield. Change the operating parameters. The change of the operation parameter for avoiding the dangerous state will be described later with reference to FIG.

  The operation instruction unit 105 outputs an operation instruction that instructs the operation of the plant equipment P0 based on the operation parameter determined by the determination unit 104. The operation instruction unit 105 outputs an operation instruction to the field device or the field operator. For example, the operation instruction unit 105 outputs a current value of 4 to 20 mA to the operation control device 4 according to the valve openings of the valves V1 to V3 determined by the determination unit 104. Further, the operation instruction unit 105 transmits an operation instruction mail to the field operator with the valve openings of the valves V1 to V3 determined by the determination unit 104 as mail text. The output of the driving instruction is executed at a predetermined timing. For example, the output of the operation instruction to the operation control device 4 is executed immediately when the operation parameter is determined. On the other hand, the operation instruction to the field operator is executed at a predetermined transmission timing so that a large amount of mail is not transmitted.

  When the operation instruction unit 105 outputs an operation instruction to the field device or the field operator, the load on the board operator can be reduced or eliminated. The operation instruction unit 105 may display the content of the operation instruction output to the field device or the field operator on the display device 15 in FIG. 2 to notify the board operator. When there is a problem with the notified driving instruction, the board operator may instruct the driving instruction unit 105 to change or cancel the driving instruction from the input device 16.

  The target yield acquisition unit 106 acquires the target yield of the product of the plant equipment P0 and inputs it to the determination unit 104. The target yield acquisition unit 106 acquires the target yield to be acquired together with the target yield setting period. The target yield acquisition unit 106 acquires, for example, a ratio (%) that is input from the input unit in FIG. 2 and sets the maximum yield in the plant P0 as 100% as the target yield. Note that the target yield acquired by the target yield acquisition unit 106 may be set in terms of quantity, volume, weight, or the like. In addition, the target yield acquisition unit 106 acquires, for example, a one-hour unit period input from the input unit in FIG. 2 as a target yield setting period.

  In FIG. 3, a case where the functions of the simulated dangerous state registration unit 101, the driving state acquisition unit 102, the learning unit 103, the determination unit 104, the driving instruction unit 105, and the target yield acquisition unit 106 are realized by software is described. did. However, one or more of the above functions may be realized by hardware. Each of the above functions may be implemented by dividing one function into a plurality of functions. Further, each of the above functions may be implemented by integrating two or more functions into one function.

  Next, the simulated dangerous state learned by the plant control apparatus 1 will be described with reference to FIG. FIG. 4 is a diagram illustrating an example of a simulated dangerous state learned by the plant control device 1 in the embodiment. The simulated dangerous state shown in FIG. 4 is registered from the simulated dangerous state registration unit 101 of FIG.

  In FIG. 4, each of KJ1 to KJ3 is one simulated dangerous state. Each simulated dangerous state is set as a numerical index in which the numerical ranges of the sensors S1 to S3 and the valves V1 to V2 of the field device are represented by a loss function. For example, the simulated dangerous state KJ1 is represented by numerical indices a11 to a19. The loss function of the numerical indices a11 to a19 may be set, for example, by a threshold value of a measured value or a mathematical formula based on the measured value.

For example, when the loss function is set in the numerical range of one temperature sensor,
Less than 50 ° C → Loss function = 0
50 ° C or higher and lower than 100 ° C → Loss function = -50
100 ° C or higher → Loss function = 0
A numerical index may be set as follows. In addition, at a temperature t ° C.
Loss function = −A1 · t + A2 (A1 and A2 are constants)
A numerical index may be set as follows.

In addition, when the loss function is set in the numerical range of the three temperature sensors (ta ° C., tb ° C. and tc ° C.),
Loss function = −B1 · ta−B2 · tb ² −B3 · tc ² (B1, B2 and B3 are constants)
A numerical index may be set as follows.

  The numerical index of the sensor S6 is a measured value obtained by measuring the yield of the product of the plant equipment P0. In the present embodiment, the yield of the product is used as one of indexes indicating the operation state of the plant. As the numerical index of the sensor S6, the loss function is set larger as the difference from the target yield acquired by the target yield acquisition unit 106 is larger. By calculating the loss function for each target yield, it is possible to calculate the loss function separately when the target yield is different. The learning unit 103 generates the operation model so that the yield is close to the target yield. For example, when the target yield is not set, the learning model 103 may generate the operation model so that the loss function is reduced as the yield increases. Good.

  The numerical index may be set by a combination of the measured values of the sensors (S1 to S6) and the opening degrees of the valves (V1 to V3). In the present embodiment, the operation state of the plant at that time is calculated as a loss function based on the measured values of the sensors S1 to S3 of the field device and the opening degrees of the valves V1 and V2. In the simulated dangerous state shown in FIG. 4, a state that is difficult to occur in the normal operation state of the plant equipment P0 is set in advance as a simulated dangerous state, and the negative bias of the loss function in the simulated dangerous state is set high. Therefore, it is possible to determine the operation parameters so as to avoid this state.

  In addition, since the simulated dangerous state can set independent conditions as shown in Kj1 to KJ3, it becomes easy to add or delete the simulated dangerous state. Further, a plurality of simulated dangerous states may overlap each other. When simulated dangerous conditions overlap, a priority order for calculating the loss function in which simulated dangerous condition may be set.

  Next, the history data learned by the plant control apparatus 1 will be described with reference to FIG. FIG. 5 is a diagram illustrating an example of history data learned by the plant control device 1 in the embodiment. The history data shown in FIG. 5 is acquired from the field device and recorded by the operation state acquisition unit 102 of FIG.

  In FIG. 5, the history data is recorded by associating the acquisition time when the driving state acquisition unit 102 acquires the measurement value with the acquired measurement value. The recording of history data is performed at regular intervals, for example. FIG. 5 shows that history data is recorded at 1-minute intervals. The record of history data may be acquired in accordance with changes in measurement values. For example, when a predetermined measurement value falls within a predetermined range, or when a change amount of the predetermined measurement value is greater than or equal to a predetermined value.

  When it is determined that the acquired history data is in a dangerous state in the operation state of the plant equipment P0, the history data may be additionally specified as a simulated dangerous state. The simulated dangerous state described with reference to FIG. 4 sets a predetermined loss function for a measured value in advance, sets a predetermined area of the measured value as a simulated dangerous state, and also measures the measured value in the same simulated dangerous state. The loss function can be changed according to the difference. On the other hand, in the history data, since the measured value at that time is acquired at one point, it is desirable that the loss function when registering as a simulated dangerous state can be individually set. In this embodiment, the case where the level of AC can be set according to the magnitude | size of a loss function is illustrated. For example, for history data acquired at 10:22 on October 10, 2016, a simulated dangerous state is added by setting an A level with a large loss function (high risk). Further, for the history data acquired at 10:23 on the same day, a simulated dangerous state is added by setting a C level with a small loss function (low risk). By setting the level of the loss function for the history data acquired in the actual operating state of the plant equipment P0 and learning it as a simulated dangerous state, it is possible to generate highly safe operating parameters.

  Next, a space including a simulated dangerous state learned by the plant control apparatus 1 will be described with reference to FIG. FIG. 6 is a diagram illustrating an example of a space including a simulated dangerous state learned by the plant control device 1 in the embodiment.

  In FIG. 6, the vertical axis and the horizontal axis indicate two-dimensional areas formed in any two numerical indexes described in FIG. 4. In FIG. 4, a total of nine numerical indices S1 to S6 and V1 to V3 are illustrated, but a space in which all nine numerical indices are combined forms a nine-dimensional space. In FIG. 6, a two-dimensional region formed by two numerical indexes is shown for the sake of simplicity.

  KJ1, KJ2, and KJ3 indicate simulated dangerous state areas set in the numerical index. That is, the operation parameter for controlling the operation state of the plant is determined so that the operation state does not enter the region of KJ1 to KJ3.

  L1, M1 and H1 indicate operating states based on operating parameters determined for each target yield. For example, when the target yield is 50%, based on the operation model generated by machine learning, the operation parameters are determined so that the yield is 50% and the L1 operation state is stabilized. The broken line region shown in the figure is a region where the yield is expected to be about 50% in the operation model generated from the history data.

  Further, when the target yield is 80%, based on the operation model generated by machine learning, the operation parameters are determined so that the operation state of M1 is stabilized at a yield of 80%. The broken line region shown in the figure is a region where the yield is expected to be about 80% in the operation model generated from the history data. When the target yield is changed from 50% to 80%, the operating state is linearly changed from L1 to M1.

  In FIG. 6, a broken-line region illustrated as the maximum yield prediction region 1 is a region where the yield obtained by machine learning is expected to be maximum. Let the operating state at this time be H1. The driving state H1 is determined so that the driving state does not become simulated dangerous states KJ1 to KJ3. The illustrated e11 to e13 indicate the size of the safety margin from the driving state H1 to each simulated dangerous state. By setting a simulated dangerous state in advance in machine learning, it is possible to predict a region where the yield is maximized safely. When the target yield is changed from 80% to 100%, the operating state is linearly changed from M1 to H1.

  In FIG. 6, a broken-line region illustrated as the yield maximum prediction region 2 is a region different from the yield maximum prediction region 1 in which the yield obtained by machine learning is predicted to be maximum. Let the operating state at this time be H2. E22 to e23 shown in the figure indicate the size of the safety margin from the driving state H2 to each simulated dangerous state.

  When the board operator instructs the field operator to operate, even if the board operator maximizes the yield, the board operator will operate in such a way that the plant will be in an operating state with some safety margin so that it will not be in a hazardous state. Specify parameters. On the other hand, in the machine learning in the present embodiment, it is possible to change the driving state while avoiding the dangerous state by setting the simulated dangerous region together with the history data. Therefore, for example, there is a possibility that the operation state of H2, which cannot change the operation state linearly from the operation state of H1, can be found by machine learning.

  Further, when the operation state is changed linearly from H1 to H2, it passes through the simulated dangerous state KJ2 or KJ3, so that it is difficult to change when the board operator gives an operation instruction to the field operator. Become. On the other hand, in the present embodiment, by setting the simulated dangerous state in advance, by appropriately determining the operating parameters that transition in the operating state that secures the safety margin calculated by the loss function from the simulated dangerous state KJ3. As shown by the arrow that transitions from H1 to H2 in FIG. 6, it is possible to change the operating state to H2 while avoiding the simulated dangerous state KJ3.

  In other words, in the present embodiment, it is possible to determine the driving parameters for the static driving states shown in H1 and H2 by creating the driving model in which the simulated dangerous state is learned in advance. In addition, it is possible to determine an operation parameter that becomes a dynamic operation state, such as the operation state transition method.

  Next, operation | movement of the plant control apparatus 1 is demonstrated using FIG. FIG. 7 is a flowchart illustrating an example of the operation of the plant control device 1 in the embodiment. The flowchart of FIG. 7 illustrates the operation of the plant control apparatus 1 when controlling the plant equipment P0 using an operation model generated in advance by machine learning.

  The operation of the flowchart of FIG. 7 is realized by the CPU 11 executing the plant control device program stored in the RMA 12 of FIG. 7 is realized by each function of the software of the plant control apparatus 1 described in FIG. In the following description, the operation of the flowchart will be described as being executed by the plant control device 1.

  In FIG. 7, the plant control apparatus 1 determines whether or not the target yield has been changed (step S11). Whether or not the target yield has been changed can be determined based on whether or not the target yield acquired by the target yield acquisition unit 106 has been changed. When it is determined that the target yield has been changed (step S11: YES), the plant control device 1 changes the set target yield (step S12). On the other hand, when it is determined that the target yield has not been changed (step S11: NO), the plant control device 1 skips the process of step S12. By skipping the process of step S12, the set target yield is maintained.

  After executing the process of step S12 or when determining that the target yield has not been changed in the process of step S11, the plant control device 1 determines whether or not the operating state has been acquired (step S13). Whether or not the operating state has been acquired can be determined by whether or not the operating state acquiring unit 102 has acquired a measurement value from the field device of the plant P0. When it is determined that the operating state has not been acquired (step S13: NO), the plant control device 1 returns to the process of step S11 and waits for the operating state to be acquired.

  On the other hand, when it is determined that the operating state has been acquired (step S13: YES), the plant control device 1 determines the operating state (step S14). In the determination of the operation state, the determination unit 104 determines the acquired operation state based on the operation model generated by the learning unit 103, and determines the operation parameter of the plant P0.

  After performing the process of step S14, the plant control apparatus 1 determines whether or not to output an operation instruction (step S15). Whether or not to output a driving instruction is determined when the driving instruction is for the driving control device 4, the driving instruction unit 105 immediately outputs the driving instruction. On the other hand, when the driving instruction is for the field operator, the driving instruction unit 105 determines that the driving instruction is output at a predetermined output timing. By changing the output timing of the driving instruction depending on the transmission destination, it is possible to output an appropriate driving instruction. When it is determined that the operation instruction is not output (step S15: NO), the plant control device 1 repeats the process of step S15 and waits for the output timing.

  On the other hand, when it is determined that an operation instruction is to be output (step S15: YES), the plant control device 1 outputs an operation instruction (step S16). The driving instruction is output in the output form corresponding to the output destination in the driving instruction unit 105.

  After executing the process of step S16, the plant control device 1 determines whether or not to end the plant control (step S17). Whether or not to end the plant control can be determined based on whether or not a stop operation of the plant control is executed from the input device 16 or the like, for example. If it is determined not to end the plant control (step S17: NO), the plant control device 1 returns to the process of step S11 and repeatedly executes the processes of step S11 to step S17. On the other hand, when it is determined that the plant control is to be ended (step S17: YES), the plant control device 1 ends the operation of the flowchart shown in FIG.

  Next, the target yield setting screen of the plant control apparatus 1 will be described with reference to FIG. FIG. 8 is a diagram illustrating an example of a target yield setting screen of the plant control apparatus in the embodiment. The target yield setting screen shown in FIG. 8 is a setting screen displayed on the display device 15 of FIG. 2 and set by the input device 16, for example.

  In FIG. 8, the target yield setting screen 1000 has a simulated dangerous state setting tab 1100, a target yield setting tab 1200, a driving instruction output setting tab 1300, and a learning setting tab 1400 at the bottom of the screen. By selecting one of these tabs, the display screen can be switched to each setting screen. A target yield setting screen 1000 shown in FIG. 8 shows a state where the target yield setting tab 1200 is selected. The simulated dangerous state setting tab 1100 is a tab for selecting a setting screen for setting the simulated dangerous state described with reference to FIG. The driving instruction output setting tab 1300 is a tab for selecting a setting screen for setting an output destination and an output form of the driving instruction. The learning setting tab 1400 is a tab for selecting a setting screen for setting the measurement value acquired by the driving state acquisition unit 102 and the acquisition timing.

  The target yield setting screen 1000 includes a target yield setting unit 1201, a yield setting period setting unit 1202, a setting button 1203, and a reset button 1204.

  The target yield setting unit 1201 has a slide bar that slides between 0% and 100%. For example, the target yield is set to a desired numerical value by operating the slide bar with a mouse. The set target yield is displayed numerically at the top of the slide bar. FIG. 8 shows that the target yield is set to 65%.

  The yield setting period setting unit 1202 sets the yield setting period for the target yield set in the target yield setting unit 1201. Similar to the target yield setting unit 1201, the yield setting period setting unit 1202 sets the yield setting period to a desired numerical value by operating a slide bar that slides between 1 hour and 90 days. For example, when the yield is increased in the short term, the yield setting period is set short. If the yield setting period is shortened, the operation parameters for operating the plant equipment P0 so that the target yield set by the target yield setting unit 1201 can be obtained at a load with a high probability that no failure or the like will occur in the plant equipment P0 because it is short. It is determined. On the other hand, when the yield setting period is lengthened, the operating parameters for operating the plant equipment P0 are determined so that the target yield set by the target yield setting unit 1201 can be obtained in a long-term stable operating state. FIG. 8 shows that the yield setting period is set to 00 hours on the 5th.

  The setting button 1203 is a button for confirming and registering the target yield and the yield setting period set by the target yield setting unit 1201 and the yield setting period setting unit 1202. A reset button 1204 is a button for resetting the target yield and the yield setting period set by the target yield setting unit 1201 and the yield setting period setting unit 1202 to initial values.

  The target yield setting screen 1000 has an operation state setting unit 1250. The driving state setting unit 1250 includes a driving instruction change button 1251, a dangerous state registration button 1252, and a driving parameter display unit 1253.

  As an auxiliary function, the operation instruction change button 1251 is a button for changing the operation parameter determined by the determination unit 104 in FIG. 2 or the operation instruction output by the operation instruction unit 105. For example, when the board operator recognizes the operation parameter displayed on the operation parameter display unit 1253 and determines that the operation parameter determined by the determination unit 104 is defective, the board operator operates the operation instruction change button 1251. It is possible to change the operating parameters determined by this. In machine learning, there are cases where an operation parameter that is an unknown operation state that is not set as a simulated dangerous state is determined, and the board operator may find a problem with the determined operation parameter. By providing the operation instruction change button 1251, the board operator can make up for inadequate machine learning.

  The dangerous state registration button 1252 is a button for adding the simulated dangerous state described with reference to FIG. When the board operator determines that the operation parameter displayed on the operation parameter display unit 1253 is to put the plant device P0 in a dangerous state, the board operator operates the dangerous state registration button 1252 to operate the operation state based on the operation parameter. Can be learned as a simulated dangerous state. Assume that the levels A to C described in FIG. 5 can be set for the simulated dangerous state to be added.

  As described above, the plant control apparatus according to the present embodiment includes a simulated dangerous state registration unit that registers a simulated dangerous state that simulates an operational state in which the plant is dangerous, and an operation that acquires the operational state of the plant. A learning unit for learning a state acquisition unit, the acquired operation state and the registered simulated dangerous state, and generating an operation model of the plant, and the acquired operation state and the generated operation model A determination unit that determines an operation parameter of the plant, and an operation instruction unit that instructs the operation of the plant based on the determined operation parameter. A control device can be provided.

  In addition, the plant control apparatus 1 mentioned above should just be an apparatus which has the function mentioned above, for example, it is implement | achieved by the system which was comprised by the combination of the some apparatus and connected each apparatus so that communication was possible. Also good. Moreover, the plant control apparatus 1 may be implement | achieved as a part of function of the operation control apparatus 4, the maintenance apparatus 6, or the field operator terminal 7 demonstrated in FIG. Note that the manufacturing execution system 3 and the operation control device 4 may be realized by a system configured by combining a plurality of devices and communicably connecting the devices, like the plant control device 1. Good.

  Further, the plant control method of the present embodiment includes an operation state acquisition step for acquiring the operation state of the plant, and a simulated danger state registration step for registering a simulated danger state that represents the operation state that makes the plant dangerous. And learning step of learning the acquired operation state and the registered simulated dangerous state to generate an operation model of the plant, and based on the acquired operation state and the generated operation model A plant control method capable of reducing the cost with high safety by including a determination step for determining an operation parameter of the plant and an operation instruction step for instructing the operation of the plant based on the determined operation parameter. Can be provided.

  In addition, the execution order of each said step in the plant control method of this embodiment is not limited to the description order of the said step, You may perform in arbitrary orders.

  Further, a program for realizing the functions constituting the apparatus described in the present embodiment is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed. Thus, the various processes described above in the present embodiment may be performed. Here, the “computer system” may include an OS and hardware such as peripheral devices. Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. The “computer-readable recording medium” means a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a CD-ROM, a hard disk built in a computer system, etc. This is a storage device.

  Further, the “computer-readable recording medium” refers to a volatile memory (for example, DRAM (Dynamic) in a computer system serving as a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. Random Access Memory)) that holds a program for a certain period of time is also included. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what implement | achieves the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

  The embodiment of the present invention has been described above with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes various modifications within the scope of the present invention. It is.

DESCRIPTION OF SYMBOLS 1 Plant control apparatus 2 Core business system 3 Manufacturing execution system 4 Operation control apparatus 5 Operation panel 6 Maintenance equipment 7 Field operator terminal 11 CPU
12 RAM
13 ROM
14 HDD
15 Display device 16 Input device 17 Communication I / F
19 Bus 101 Simulated dangerous state registration unit 102 Driving state acquisition unit 103 Learning unit 104 Determination unit 105 Driving instruction unit 106 Target yield acquisition unit

Claims (10)

A simulated dangerous state registration unit for registering a simulated dangerous state that simulates the operating state in which the plant is dangerous;
An operation state acquisition unit for acquiring the operation state of the plant;
A learning unit that learns the acquired operation state and the registered simulated dangerous state, and generates an operation model of the plant;
A determination unit that determines an operation parameter of the plant based on the acquired operation state and the generated operation model;
Bei example a driving instruction unit based on the determined the operating parameters for instructing the operation of the plant,
The learning unit learns the acquired driving state and the registered simulated dangerous state as a loss function indicating incompatibility of the driving state, and associates the driving state with the loss function to associate the driving model. Produces
The said determination part is a plant control apparatus which calculates the said loss function of the acquired said operation state, and determines the said operation parameter which decreases the said loss function based on the produced | generated operation model . The plant control device according to claim 1 , wherein the learning unit generates the operation model such that the loss function increases in the learned simulated dangerous state. The plant control device according to claim 1 , wherein the operation instruction unit instructs the device that adjusts the operation state to operate the plant. The plant control device according to any one of claims 1 to 3 , wherein the operation instruction unit instructs an operator who operates an apparatus for adjusting the plant to operate the plant. A target yield acquisition unit for acquiring the target yield of the plant;
The said determination part determines the said operation parameter so that the said simulated dangerous state may be avoided based on the said driving model, when the acquired said target yield is changed, The any one of Claim 1 to 4 The plant control apparatus according to item. The target yield acquisition unit further acquires a set period of the target yield,
The plant control device according to claim 5 , wherein the determination unit determines the operation parameter based on the acquired target harvest amount in the set period. An operation control unit for controlling plant equipment;
A manufacturing execution unit for managing the operating state of the device;
A simulated dangerous state registration unit for registering a simulated dangerous state that simulates an operational state in which the plant is dangerous;
An operation state acquisition unit for acquiring the operation state of the plant;
A learning unit that learns the acquired operation state and the registered simulated dangerous state, and generates an operation model of the plant;
A determination unit that determines an operation parameter of the plant based on the acquired operation state and the generated operation model;
E Bei the operation instruction section for instructing the operation of the plant with respect to the operation control unit based on the determined the operating parameters,
The learning unit learns the acquired driving state and the registered simulated dangerous state as a loss function indicating incompatibility of the driving state, and associates the driving state with the loss function to associate the driving model. Produces
The said determination part is a plant control apparatus which calculates the said loss function of the acquired said operation state, and determines the said operation parameter which decreases the said loss function based on the produced | generated operation model . A simulated dangerous state registration step for registering a simulated dangerous state that simulates an operational state in which the plant is dangerous;
An operation state acquisition step of acquiring the operation state of the plant;
A learning step of learning the acquired operating state and the registered simulated dangerous state, and generating an operating model of the plant,
A determination step for determining an operation parameter of the plant based on the acquired operation state and the generated operation model;
Based on the determined the operating parameters see contains an operation instruction step for instructing the operation of the plant,
The learning step learns the acquired driving state and the registered simulated dangerous state as a loss function indicating incompatibility of the driving state, and associates the driving state with the loss function to associate the driving model. Produces
The plant control method , wherein the determining step calculates the loss function of the acquired operating state, and determines the operating parameter for decreasing the loss function based on the generated operating model . A simulated dangerous state registration process for registering a simulated dangerous state that simulates the operating state in which the plant is dangerous,
An operation state acquisition process for acquiring the operation state of the plant;
A learning process of learning the acquired operating state and the registered simulated dangerous state, and generating an operating model of the plant,
A determination process for determining operating parameters of the plant based on the acquired operating state and the generated operating model;
A plant control program for causing a computer to execute an operation instruction process for instructing operation of the plant based on the determined operation parameter,
The learning process learns the acquired driving state and the registered simulated dangerous state as a loss function indicating incompatibility of the driving state, and associates the driving state with the loss function to associate the driving model. Produces
The said determination process is a plant control program which calculates the said loss function of the acquired said operation state, and determines the said operation parameter which decreases the said loss function based on the produced | generated operation model . A simulated dangerous state registration process for registering a simulated dangerous state that simulates the operating state in which the plant is dangerous,
An operation state acquisition process for acquiring the operation state of the plant;
A learning process of learning the acquired operating state and the registered simulated dangerous state, and generating an operating model of the plant,
A determination process for determining operating parameters of the plant based on the acquired operating state and the generated operating model;
A recording medium recording a plant control program for causing a computer to execute an operation instruction process for instructing operation of the plant based on the determined operation parameter ,
The learning process learns the acquired driving state and the registered simulated dangerous state as a loss function indicating incompatibility of the driving state, and associates the driving state with the loss function to associate the driving model. Produces
The determination process calculates the loss function of the acquired operation state, and determines the operation parameter for decreasing the loss function based on the generated operation model .

Images